Memory systems for computers provide many memory devices on a common bus to allow larger storage and transmission capacities than can be obtained with a single memory device. The memory devices are multiplexed on to a multidrop bus to reduce the pin count of a memory bus master or controller. Most of these systems require user upgradeable or replaceable components to allow future expansion or repair of the memory subsystems. Typically, these systems are upgraded on a module basis, where the memory module (e.g., a dual in-line memory module or DIMM) has several devices on a small printed circuit board (PCB), and the module plugs into a connector that provides an electrical connection to the memory subsystem bus.
From a signal integrity standpoint, the provision of many memory devices on the bus can be problematic since these modules represent electrical stubs to the memory bus, which causes reflection on the bus. These reflections degrade the signal integrity and therefore, limit the maximum bandwidth or timing margin of the system. A robust electrical design is required in a high speed multidrop memory bus since the signal integrity must be acceptable to lightly loaded systems, that is, where only a small number of module slots are populated, heavily loaded systems, and for every device on the bus. A signal analysis of a typical memory subsystem has shown degraded signal integrity when the memory subsystem is fully loaded.
An example of a multidrop memory bus that must carefully balance the design for different loading characteristics is one which is intended for use with a double data rate synchronous dynamic random access memory (DDR SDRAM) main memory system. Such systems often have up to four memory slots that operate at a bus frequency of at least 133 MHz. Each memory slot can be populated with a single bank or double bank memory module. Balancing the design to be acceptable for both lightly and fully loaded situations can be challenging due to the number of slots, varying number of banks on the memory modules, and minor impedance mismatches between the memory modules and the memory bus.
Now referring to the drawings, where like reference numerals designate like elements, there is shown in FIG. 1 a conventional memory system 1. The memory system 1 includes a memory controller 200, which may be coupled to a computer system via a local bus 1000, which is also coupled to a processor 1100 and an expansion bus controller 1200. The expansion bus controller 1200 is also coupled to one or more expansion buses 2000, to which various peripheral devices such as mass storage devices, keyboard, mouse, graphics adapters, and multimedia adapters may be attached.
The memory controller 200 is also coupled to a memory bus 100, which includes a plurality of sockets 106a-106d. The sockets 106a-106d may be left empty, or they can accept memory modules 300a-300d. The memory modules may be double bank modules containing a first memory bank 301a-301d and a second memory bank 302a-302d, respectively, or the memory modules may be single banked modules containing only the first memory bank 301a-301d. 
In order to operate the memory bus 100 at high speed, it is important to minimize signal reflections within the bus. To this end, the memory bus 100 includes a transmission line 101 that contains a source resistor 105, which splits the transmission line 101 into a first segment 102 running from the memory controller to the source resistor 105 and a second segment 103 which runs from the source resistor 105 to a terminator 104 and which includes the plurality of sockets 106a-106d. The terminator 104 includes a terminating resistor Rterm and a termination voltage source VTT. The use of the source resistor 105, terminating resistor Rterm and termination voltage source VTT is designed to match the memory bus 100 loaded impedance. When the memory bus is populated with memory modules 300a-300d (via the sockets 106a-106d), electrical stubs are created on the memory bus. These stubs reduce the effective impedance at that point on the bus, and this creates signal reflections which reduce the signal integrity and the maximum possible data rate that can be transferred on the bus.
When a four socket memory system has each socket populated by a double bank memory module, there are a large number of minor impedance mismatches leading to a significant decrease in signal integrity. FIG. 2A-2D are examples of signal plots of read operations from each of the four double bank memory modules 300a-300d, respectively. Similarly, FIGS. 3A-3D are examples of signal plots of write operations to each of the four double bank memory modules 300a-300d, respectively.
Each signal plot shows a reference voltage 10, an aperture box 20 for a first overdrive voltage, and an aperture box 30 for a second overdrive voltage. The reference voltage 10 is the baseline voltage of the memory bus 100. Signals are detected on the memory bus 100 by either the memory controller 200 or the memory modules 300a-300d when the voltage level of the signal differs by a minimum threshold, or overdrive voltage threshold, from the reference voltage 10. For example, a logical low, sometimes called voltage output low or VOL is detected on the memory bus 100 when the signal is at a voltage below the difference between the reference voltage 10 and the overdrive threshold voltage, while a logical high, sometimes called voltage output high or VOH is detected when the signal is at a voltage above the sum of the reference voltage 10 and the overdrive voltage. Two separate overdrive voltage thresholds are shown on the signal plots because differing memory systems may require different overdrive thresholds. For example, the use of the larger second overdrive parameter may result in more accurate signal detection in a noisy environment. The two aperture boxes 10, 20 illustrate the period of time when the plotted signals 40 differed by at least a first or second overdrive voltage threshold, respectively, to be detectable as either voltage output high or voltage output low. The plotted signals 40 are the signals that are seen by the memory controller 200 when the memory modules 300a-300d drive signals onto the memory bus 100 (i.e., for the read operations illustrated in FIGS. 2A-2D), as well as the signals seen at each memory module 300a-300d when the memory controller 200 drives signals onto the memory bus 100 (i.e., for the write operations illustrated in FIGS. 3A-3D). In each case, the signals driven onto the memory bus 100 are a plurality of pseudo-random pulses.
As illustrated in FIGS. 2A-2D and FIGS. 3A-3D, the conventional system exhibits the following characteristics. When using the first overdrive threshold of 0.31 volts for read operations, the four memory modules have signal aperture times of 2.33 nanoseconds (ns), 2.29 ns, 2.33 ns, and 2.29 ns, respectively. For writes, the aperture times are 1.25 ns, 1.67 ns, 1.83 ns, and 1.92 ns, respectively. When using the second (larger) overdrive voltage threshold of 0.35 volts for read operations, the four memory modules have aperture times of 0.83 ns, 1.83 ns, 2.04 ns, and 2.00 ns, respectively. For writes, the aperture times are 0.71 ns, 1.25 ns, 1.54 ns, 1.58 ns. Thus, a fully loaded conventional memory bus 100 exhibits poor aperture times for write operations, especially when the overdrive threshold is set at 0.35 volts. Additionally, reads from the first memory module also exhibit poor aperture times at the 0.35 volt overdrive threshold.
Accordingly, there is a desire and need to improve the signal integrity of a fully loaded memory system in order to permit high speed operation.